Fast UAV Object-Searching in Large-Scale and Complex Environments

Abstract

Autonomous object-searching is crucial for various applications of unmanned aerial vehicles (UAVs). Considering the fact that existing autonomous exploration methods either focus only on maximizing the exploration of unknown areas or suffer from insufficient searches due to repeated and unnecessary exploration, this article introduces an effective object-searching strategy for UAVs in large-scale and complex environments. A novel method is proposed to empower UAVs with the capability to conduct fast, secure, and efficient searches for interested objects in large-scale and complex environments. A Kalman filter-based YOLO algorithm is first proposed to achieve robust object position estimation in cluttered and occlusion-prone scenarios, and a mode-based method is then introduced to conduct a computationally efficient viewpoint generation. A hierarchical searching method is proposed, which not only can increase computational and search efficiency but also can leverage frontier data for search-planning, including coarse global searching paths and optimizing local refined searching trajectories. Experimental results in six different environments indicate that our proposed method outperforms existing techniques in terms of both reduced searching times and computing time. Moreover, the effectiveness of the proposed method is substantiated in various real-world scenarios.